Re: [DynInst_API:] [PATCH] proccontrol: Avoid a destructor race on Counter::locks

[Bill Williams <bill@xxxxxxxxxxx>]

On 05/16/2014 07:01 PM, Josh Stone wrote:
> On 05/16/2014 04:03 PM, Josh Stone wrote:
> > +Mutex<false> * Counter::locks = new Mutex<false>[Counter::NumCounterTypes];
> >   int Counter::global_counts[Counter::NumCounterTypes];
> FWIW, I considered ditching the mutexes and making global_counts[] use
> atomic<int>, but we don't yet live in a world where that is portable.
>
> RHEL6 doesn't have boost::atomic.  It does have std::atomic, but it's
> found in <cstdatomic> rather than <atomic>.  On Windows, the builders
> are using VS10, which doesn't have std::atomic.  At least there we could
> use new enough boost for boost::atomic.  Or either could use intrinsics,
> via GCC __sync_fetch_and_add or Windows InterlockedAdd, but that's not
> even trying to be portable.
>
> So rather than a bunch of platform ifdefs, I chose to just make the
> existing mutexes work more consistently.
> _______________________________________________
> Dyninst-api mailing list
> Dyninst-api@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/dyninst-api
Atomics would be delightful, but I agree that now is not the time.

We could of course replace the array of locks with a 
vector<Lock>-derived class whose sole purpose is to have its destructor 
wait for the process to become single-threaded again, but really, 
leaking at exit should be harmless.
--
--bw

Bill Williams
Paradyn Project
bill@xxxxxxxxxxx